System and method for identification of colors in digital images

Abstract

An aspect of some embodiments of the present invention relates to a method for comparing a color of a colored region to a plurality of predetermined discrete colors, the method comprising: a) identifying, on a digital image, a colored region; b) determining a color of the colored region, and determining a set of first values of color components forming the color, the color components belonging to a predetermined color space; c) receiving parameters from a machine learning system trained to identify a plurality of color groups based on training data, and each color group corresponding to a respective discrete color of a plurality of predetermined discrete colors, the parameters being indicative of the color groups; d) processing the first values with at least some of the parameters in order to associate the color of the colored region with one of the predetermined discrete colors.

What is claimed is: 
     
         1 . A method for comparing a color of a colored region to a plurality of predetermined discrete colors, the method comprising:
 a) identifying, on a digital image, a colored region;   b) determining a color of the colored region, and determining a set of first values of color components forming the color, the color components belonging to a predetermined color space;   c) receiving parameters from a machine learning system trained to identify a plurality of color groups based on training data, and each color group corresponding to a respective discrete color of a plurality of predetermined discrete colors, the parameters being indicative of the color groups; and   d) processing the first values with at least some of the parameters in order to associate the color of the colored region with one of the predetermined discrete colors;   wherein (b) determining a color of the colored region comprises:   providing a plurality of positions in the colored region;
 determining, for each position of the plurality of positions, a respective set of second values of the color components belonging to the predetermined color space; and 
 averaging the second values of each color component of the plurality of positions, to yield a set of average values of the color components, the set of average values of the color components being the set of first values of color components of the color of the colored region. 
   
     
     
         2 . The method of  claim 1 , wherein (a) providing, on a digital image, a colored region comprises:
 providing a location pixel indicative of a location of the colored region.   
     
     
         3 . The method of  claim 1 , wherein the predetermined color space is an RGB color space, and the color components are a red color component, a green color component, and a blue color component. 
     
     
         4 . The method of  claim 1 , comprising, prior to step (a):
 placing a test strip having the colored region on a recessed surface surrounded by a line;   providing a raw digital image of the test strip in the recessed surface;   detecting the line;   obtaining the digital image by processing the raw digital image, by keeping from the raw digital image of the test strip in the recessed surface only a section that is within the line surrounding the recessed surface.   
     
     
         5 . The method of  claim 1 , comprising, prior to step (a):
 placing a test strip having the colored region on a surface;   tilting the surface with respect to a horizontal plane;   capturing a photograph of the test strip while tilted;   processing the photograph to provide the digital image.   
     
     
         6 . A method for training a machine learning system to generate the plurality of color groups, the method comprising:
 providing training data in form of a plurality of first digital images having respective first colored regions having respective first colors and providing a known discrete color of the plurality of predetermined discrete colors for each first color;   generating training data points for a plurality of colors assumed by the first colored regions by using the first digital images, each training data point corresponding to a point in a predetermined color space;   instructing the machine learning system to detect clusters of the training data points and to create partitions between the clusters; and   outputting parameters of the partitions;   
       wherein each training data point corresponds to a respective first colored region of a respective first digital image, and each training point being determined by detecting a training value of each of the color components belonging to the predetermined color space in the first colored region. 
     
     
         7 . The method of  claim 6 , wherein each the training value is calculated by:
 detecting first-region values of the color components belonging to the predetermined color space at a plurality of spots in the respective first colored region;   for each of the color components averaging the first-region values of the spots, thereby generating the training value of each color of the components.   
     
     
         8 . A method for analysis of a digital image of one or more reagent pads on a test strip, each of the one or more reagent pads being configured to assume one of a respective plurality of colors upon contacting a liquid, depending on a quantity of a substance in the liquid, the method comprising:
 a) providing, on a digital image of a test strip, a set of location pixels, each location pixel indicative of a location of a respective one of one or more reagent pads of the test strip;   b) determining, for a plurality of positions around the location pixel of each of the one or more reagent pads, values of color components belonging to a predetermined color space;   c) for each one of the one or more reagent pads, separately averaging the values of each of the color components at the positions, thereby generating an average input value for each of the color components and generating a color of the pad as a set comprising the average input values;   d) for each one of the one or more reagent pads, providing a respective set of parameters from a machine learning system trained to identify a plurality of color groups which correspond to respective diagnoses for a substance reacting with the one of the reagent pad, the parameters being indicative of the color groups;   e) for each one of the one or more reagent pads, processing the average input values of the color components of the reagent pad with at least some of the parameters indicative of the color groups associated with the reagent pad, to determine which of the color groups the color of the reagent pad fits into, thereby associating the color of the reagent pad with a diagnosis of the one of color groups corresponding the reagent pad, and outputting a diagnosis relating to the substance reacting with the one of the one or more reagent pads.   
     
     
         9 . A method for comparing a color of a colored region to a plurality of predetermined discrete colors, the method comprising:
 a) identifying, on a digital image, a colored region;   b) determining a color of the colored region, and determining a set of first values of color components forming the color, the color components belonging to a predetermined color space;   c) receiving parameters from a machine learning system trained to identify a plurality of color groups based on training data, and each color group corresponding to a respective discrete color of a plurality of predetermined discrete colors, the parameters being indicative of the color groups; and   d) processing the first values with at least some of the parameters in order to associate the color of the colored region with one of the predetermined discrete colors;   wherein (a) providing, on a digital image, a colored region comprises providing a location pixel indicative of a location of the colored region; and   wherein (b) determining a color of the colored region comprises:
 providing a plurality of positions within a predetermined distance from the location pixel; 
 determining, for each position in the plurality of positions, a respective set of second values of the color components belonging to the predetermined color space; and 
 averaging the second values of each color component of the plurality of positions, to yield a set of average values of the color components, the set of average values of the color components being the set of first values of color components of the color of the colored region. 
   
     
     
         10 . A method for comparing a color of a colored region to a plurality of predetermined discrete colors, the method comprising:
 a) identifying, on a digital image, a colored region;   b) determining a color of the colored region, and determining a set of first values of color components forming the color, the color components belonging to a predetermined color space;   c) receiving parameters from a machine learning system trained to identify a plurality of color groups based on training data, and each color group corresponding to a respective discrete color of a plurality of predetermined discrete colors, the parameters being indicative of the color groups; and   d) processing the first values with at least some of the parameters in order to associate the color of the colored region with one of the predetermined discrete colors;   wherein (a) providing, on a digital image, a colored region comprises providing a location pixel indicative of a location of the colored region by:
 resizing the digital image to generate a resized image of a predetermined pixel size; 
 determining, for each pixel of the resized image, a color saturation value; 
 finding a peak saturation value among the color saturation values, the peak corresponding to the location of the colored region; 
 setting a location corresponding to the peak saturation value as the location pixel of the colored region. 
   
     
     
         11 . A method for comparing a color of a colored region to a plurality of predetermined discrete colors, the method comprising:
 a) identifying, on a digital image, a colored region;   b) determining a color of the colored region, and determining a set of first values of color components forming the color, the color components belonging to a predetermined color space;   c) receiving parameters from a machine learning system trained to identify a plurality of color groups based on training data, and each color group corresponding to a respective discrete color of a plurality of predetermined discrete colors, the parameters being indicative of the color groups; and   d) processing the first values with at least some of the parameters in order to associate the color of the colored region with one of the predetermined discrete colors;   wherein the digital image is provided such that the colored region is substantially symmetrical about a vertical axis;   wherein (a) providing, on a digital image, a colored region comprises providing a location pixel indicative of a location of the colored region by:
 resizing the digital image to generate a resized image of a predetermined pixel size; 
 determining, for each pixel of the resized image, a color saturation value; 
 averaging the color saturation values horizontally, thereby determining a plurality of average horizontal saturation values; 
 finding a peak among the average horizontal saturation values; 
 setting a vertical location of the peak as a vertical coordinate of the location pixel of the colored region; 
 setting a horizontal location of the vertical axis as a horizontal coordinate of the colored region. 
   
     
     
         12 . A method for comparing a color of a colored region to a plurality of predetermined discrete colors, the method comprising:
 a) identifying, on a digital image, a colored region;   b) determining a color of the colored region, and determining a set of first values of color components forming the color, the color components belonging to a predetermined color space;   c) receiving parameters from a machine learning system trained to identify a plurality of color groups based on training data, and each color group corresponding to a respective discrete color of a plurality of predetermined discrete colors, the parameters being indicative of the color groups; and   d) processing the first values with at least some of the parameters in order to associate the color of the colored region with one of the predetermined discrete colors;   wherein:   step (a) comprises: identifying, on the digital image, a plurality of separate colored regions, each colored region being surrounded by a background region;   step (b) comprises: for each colored region, determining a respective color of the colored region, and determining a respective set of first values of color components forming the respective color, the color components belonging to the predetermined color space;   step (c) comprises: for each colored region, receiving a respective set of parameters from a machine learning system trained to generate a set of color groups, each set of color groups being associated with a respective colored region of the plurality of colored regions, each color group corresponding to a respective discrete color of a respective plurality of predetermined discrete colors associated with the colored region, the parameters in each set of parameters being indicative of the color groups; and   step (d) comprises: for each colored region, processing the first values with at least some of the parameters indicative of the color groups associated with the colored region, in order to associate the color of the colored region with one the predetermined discrete colors.   
     
     
         13 . A method for comparing a color of a colored region to a plurality of predetermined discrete colors, the method comprising:
 a) identifying, on a digital image, a colored region;   b) determining a color of the colored region, and determining a set of first values of color components forming the color, the color components belonging to a predetermined color space;   c) receiving parameters from a machine learning system trained to identify a plurality of color groups based on training data, and each color group corresponding to a respective discrete color of a plurality of predetermined discrete colors, the parameters being indicative of the color groups; and   d) processing the first values with at least some of the parameters in order to associate the color of the colored region with one of the predetermined discrete colors;   wherein the method comprises, prior to step (a), training the machine learning system to generate the plurality of color groups, by:
 providing training data in form of a plurality of first digital images having respective first colored regions having respective first colors and providing a known discrete color of the plurality of predetermined discrete colors for each first color; 
 generating training data points for a plurality of colors assumed by the first colored regions by using the first digital images, each training data point corresponding to a point in the predetermined color space; 
 instructing the machine learning system to create a predictive model by identifying clusters of the training data points in the color space, using locations of the points in the color space and the known discrete colors of the training data points, and creating partitions between the clusters; 
 wherein the parameters are parameters of the partitions separating the clusters. 
   
     
     
         14 . The method of  claim 13 , wherein each training data point corresponds to a respective first colored region of a respective first digital image, and each training point being determined by:
 detecting a training value of each of the color components belonging to the predetermined color space in first colored region.   
     
     
         15 . The method of  claim 14 , wherein each training value is calculated by:
 detecting first-region values of the color components belonging to the predetermined color space at a plurality of spots in the respective first colored region;   for each of the color components averaging the first-region values of the spots, thereby generating the training value of each color of the components.   
     
     
         16 . The method of  claim 13 , wherein the machine learning system comprises a support vector machine (SVM). 
     
     
         17 . The method of  claim 13 , wherein the generating of training data points further comprises augmenting the training data by creating one or more pseudo samples, the creating of one of the pseudo samples comprising:
 providing the discrete colors of the plurality of predetermined discrete colors, and for each of the discrete colors providing a set of predetermined color component values which form the predetermined color in the color space;   for each of the predetermined discrete colors, identifying the training data points corresponding to the predetermined discrete color and averaging the training values of each of the color components of the identified training data points, thereby creating a set of averaged training color components, and calculating a standard deviation corresponding to each of the averaged training color components;   generating a pseudo training data point for each of the predetermined discrete colors, by adding to each one of the predetermined color component values a random number ranging from a negative value of the standard deviation to a positive value of the standard deviation corresponding to a respective average training color component.   
     
     
         18 . A method for training a machine learning system to generate the plurality of color groups, the method comprising:
 providing training data in form of a plurality of first digital images having respective first colored regions having respective first colors and providing a known discrete color of the plurality of predetermined discrete colors for each first color;   generating training data points for a plurality of colors assumed by the first colored regions by using the first digital images, each training data point corresponding to a point in a predetermined color space;   instructing the machine learning system to detect clusters of the training data points and to create partitions between the clusters;   outputting parameters of the partitions;   wherein the generating of the training data points further comprises augmenting the training data by creating one or more pseudo samples, the creating of one of the pseudo sample comprising:
 providing the discrete colors of the plurality of predetermined discrete colors, and for each of the discrete colors providing a set of predetermined color component values which form the predetermined color in the color space; 
 for each of the predetermined discrete colors, identifying the training data points corresponding to the predetermined discrete color and averaging the training values of each of the color components of the identified training data points, thereby creating a set of averaged training color components, and calculating a standard deviation corresponding to the averaged training colored components; 
 generating a pseudo training data point for each of the predetermined discrete colors, by adding to each one of the predetermined color component values a random number ranging from a negative value of the standard deviation to a positive value of the standard deviation corresponding to a respective averaged training color component.